Error in running model for group convolution as" Fatal: Failed to assign a backend to each layer"

[RavikumarLav]

HI ARMNN team,

i have tried to port one of unet model where it has group convolution.But while running getting below error.

Warning: WARNING: Layer of type Convolution2d is not supported on requested backend CpuAcc for input data type Float32 and output data type Float32 (reason: in validate src/cpu/operators/CpuGemmConv2d.cpp:717: weights->dimension(idx_channel) != src->dimension(idx_channel)), falling back to the next backend.
Warning: ERROR: Layer of type Convolution2d is not supported on any preferred backend [CpuAcc ]
Fatal: Failed to assign a backend to each layer

is group convolution not supported in ARMNN?

[morgolock]

Hi @RavikumarLav

Could you please share the model with use?

[RavikumarLav]

hi @morgolock
Sorry i can't share our model here but i will share the screenshot of the layer where i am getting error.
here first convolution layer there is no problem but in the second convolution there is problem in compilation.

[morgolock]

Hi @RavikumarLav
From the picture you shared I can see the shapes contain ?, for example ?x256x256x1. It looks like the model requires dynamic shapes? ArmNN does not support dynamic shapes.

[RavikumarLav]

HI @morgolock
->From the netron app even though it is showing ?x26x256x1 but it is 1x256x256x1 that i am able to see while debugging inside computelib .In the first convolution there is no problem but at the second convolution node input is 1x256x256x32 but filter is 32x5x5x4(while generating the model in pytorch I kept group=8(8x4=32)) here it is throwing above error(since from input 32 is the channel but in filter it is 4).
->This is not working with CpuAcc but it is working with CpuRef mode.